Benchmarks/Qwen3-4B-Instruct-2507-unsloth-Q5_K/bench_metrics.json

{
  "llamabench": {
    "backend": "CUDA",
    "log_path": "Qwen3-4B-Instruct-2507-unsloth-Q5_K/llamabench.md",
    "ngl": "35",
    "raw_row": {
      "backend": "CUDA",
      "model": "qwen3 4B Q5_K - Medium",
      "ngl": "35",
      "params": "4.02 B",
      "size": "2.69 GiB",
      "t/s": "385.17 \u00b1 7.14",
      "test": "pp8",
      "tps_value": 385.17
    },
    "test": "pp8",
    "tps": 385.17
  },
  "perplexity": {
    "code": {
      "log_path": "Qwen3-4B-Instruct-2507-unsloth-Q5_K/perplexity_code.log",
      "ppl": 1.5542,
      "ppl_error": 0.01235
    },
    "general": {
      "log_path": "Qwen3-4B-Instruct-2507-unsloth-Q5_K/perplexity_general.log",
      "ppl": 8.9707,
      "ppl_error": 0.20785
    },
    "math": {
      "log_path": "Qwen3-4B-Instruct-2507-unsloth-Q5_K/perplexity_math.log",
      "ppl": 6.7701,
      "ppl_error": 0.13838
    }
  }
}